Improved detection of steganographic algorithms in spatial LSB Stego images using hybrid GRASP-BGWO optimisation

Abstract

With the success of passive steganalysis, active steganalysis proceeds with its first step to reveal the steganographic algorithms being used to create the stego images. This process needs to be modelled as a multi-class classification problem. Stem to stern analysis of the literature points out that the existing universal steganalytic features are a thorn in the flesh because of their dimensionality curse (34,671). Hence this work concentrates on detection of steganographic algorithms by optimal novel features christened Local Residual Pattern (LRP) and Local Distance Pattern (LDiP). LRP captures first order derivatives of the high pass filtered output, while LDiP exploits the multi scaled radii neighbourhood to capture deformities at a distance. Acquiring LRP and LDiP from fifteen different kernels, this work focuses to find optimal features by the proposed hybrid technique of Greedy Randomised Adaptive Search—Binary Grey Wolf Optimisation (GRASP-BGWO). Deriving confidence from the bio-inspired algorithm and the divide and conquer approach of the proposed optimisation, this work succeeds in improving the performance of the employed ensemble logistic regression classifier with minimal features. Experimentations conducted using five representative algorithms of spatial Least Significant Bit (LSB) embedding category for eight different payloads show that the developed minimal feature steganalyser outperforms the state-of-the-art steganalysers.